Claude Desktop Bridge MCP server

Claude-Desktop-Bridge

Built for Claude Desktop users. If you only use the standalone claude CLI or Claude.ai in a browser, this isn't for you — skip to Where this works for the one exception.

The problem

Claude Desktop's Chat tab and Code tab live in the same app, one click apart. But they don't share a single byte of context.

Ask Claude something in Chat, switch to Code to actually build it — Claude in Code has no memory the conversation ever happened. Spend an hour in Code working through a decision, hop to Chat to think out loud about what's next — Chat starts from zero. Two tabs, one window, and no connection between them at all.

So close, yet so far. The workaround is re-explaining yourself every time you switch tabs — burning tokens and patience re-deriving context that existed ten seconds ago, one tab over.

The solution

Claude-Desktop-Bridge is a small local MCP server, installed once into Claude Desktop, that gives Chat and Code a shared, persistent knowledge vault. Save a note while working in Code, read it back in Chat five minutes later. Research something in Chat, recall it in Code without repeating yourself.

It's deliberately not a context dump. Instead of pasting whole documents or files into the conversation, Claude gets five narrow tools — save, retrieve, search, update, delete — so it pulls in only what's actually relevant, on demand, instead of loading everything every time.

Notes can be tagged with an optional project field, so if you work across multiple codebases, the vault doesn't turn into one undifferentiated pile — each project's notes can be filtered independently.

Install

1. Download claude-desktop-bridge.mcpb from Releases.
2. Drag it onto the Claude Desktop window — or Settings → Extensions → Advanced settings → Install Extension…
3. That's it. No configuration screen, nothing to fill in. The vault lives at ~/.claude-bridge/vault.db.

Use it

Once installed, just talk normally in either tab — Claude calls savenote, getnotes, search_notes, etc. on its own when it's useful, the same way it uses any other tool. You don't invoke anything manually.

A few things worth knowing:

• It's proactive, not automatic-automatic. Claude decides when something's worth saving or worth searching for — it's not logging every message. If you want something remembered, it helps to say so ("remember this for later").
• Tag by project if you work across multiple codebases. Ask Claude to filter by project when recalling notes, so a search in one repo's context doesn't surface notes from an unrelated one.
• Chat tab "Projects" (claude.ai-style, with Custom Instructions) don't auto-link to a Code tab folder. There's no protocol-level way for Chat to know which codebase it corresponds to. If you want a Chat Project to stay in sync with a specific repo's notes, say so explicitly in that Project's Custom Instructions — e.g. "This project corresponds to the my-repo codebase; use Claude-Desktop-Bridge with project: 'my-repo'."

Where this works

Shortened. The full README is on GitHub.